JP2624880B2 - Control device and control method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a control system for controlling a control target operated by a plurality of actuators, by comprehensively determining the operation of the control target and a plurality of actuators, and The present invention relates to a control device and a control method for determining an optimal control amount of an actuator.

[Conventional technology]

The conventional control device uses a plurality of output signals indicating the operation state of the control target, and performs control.However, in a method in which control is performed by looking at individual outputs, control becomes local, for example, in a tandem rolling mill. Such a rolling system has a drawback that the optimality of the entire system cannot be considered.

Therefore, recently, there is a tendency to use a plurality of signals to obtain the overall optimality. For example, the operation of a conventional control device and a conventional method will be described using a rolling mill system that is complicated to be controlled and cannot be controlled by a single control device.

2. Description of the Related Art A rolling mill is a system for obtaining a steel material having a desired thickness by controlling the interval between opposed rolls and the tension applied to a rolled material. However, a flat steel material cannot be obtained due to thermal deformation due to heat loss generated during rolling, mechanical deformation, or the like, and therefore, shape control has been developed to obtain flat characteristics.

However, since the physical characteristics of rolling vary greatly depending on individual factors, even if a control model is created near a specific operating point and controlled, in many cases, the model will not work with the actual operation of the rolling mill. Will be different. For this reason, there is a problem that the feedback control which gives a good result if the model is accurate cannot sufficiently exhibit its ability and cannot exceed the skilled operator who operates with intuition and experience.

[Problems to be solved by the invention]

The prior art described above does not take into account the use of the expertise of a skilled operator, and has a problem in control performance.

An object of the present invention is to provide a scalable control device and control method incorporating the know-how of a skilled operator.

[Means for solving the problem]

The above object is achieved by detecting the speed of an electric motor driving a roll of a rolling mill, which is detected as a state quantity of an actuator from at least a detector provided in each rolling mill of a rolling system for rolling a material to be rolled by a plurality of rolling mills, and At least one of the temperature and tension of the material to be rolled, which is detected as a state quantity of the material to be rolled from the load of the rolling-down device for controlling the sheet thickness of the material and the state quantity of the material to be rolled from the detector provided on the outlet side of each rolling mill In the control device that controls the amount of reduction of the reduction device and the speed of the electric motor that each of the plurality of rolling mills has based on, the detected speed of the electric motor, the load of the reduction device, the temperature of the material to be rolled, and the temperature of the material to be rolled. An input unit for inputting at least one state quantity of the tension from a detector provided on each of the rolling mills or a detector provided on the output side of each rolling mill, and a state quantity input from the input unit. The stored plurality of patterns are stored in advance, and the degree of similarity between the pattern generated by combining the input state quantities along the traveling direction of the material to be rolled and the previously stored pattern is stored in the previously stored plurality of patterns. And a pattern recognition unit having an output unit that obtains each of the patterns and outputs a similarity to each of the plurality of stored patterns, and a certainty factor and a speed for each of the plurality of patterns stored in the pattern recognition unit. A membership function that defines a relationship between the pattern and a certainty factor and a reduction command pattern is stored, and the similarity output by the pattern recognition means is input, and the stored member is used as the certainty factor for the similarity. A speed pattern and a rolling command pattern are inferred from the ship function, and the inferred speed pattern and the rolling command pattern are deduced. It can be achieved by providing a command generation mechanism to determine the reduction rate of the motor speed and pressure device of each rolling mill on the basis of.

The above object is also achieved by a speed of an electric motor driving a roll of a rolling mill, which is detected as a state quantity of an actuator from a detector provided in at least each rolling mill of a rolling system for rolling a material to be rolled by a plurality of rolling mills, and At least one of the temperature and tension of the material to be rolled, which is detected as a state quantity of the material to be rolled from the load of the rolling-down device for controlling the sheet thickness of the material and a state quantity of the material to be rolled from a detector provided on the outlet side of each rolling mill. In the control device that controls the amount of reduction of the reduction device and the speed of the electric motor that each of the plurality of rolling mills has, based on the speed of the electric motor, the load of the reduction device, the temperature of the material to be rolled, and the tension of the material to be rolled. At least one state quantity is input from a detector provided in each of the rolling mills or a detector provided on the output side of each rolling mill, and a plurality of parameters generated for the input state quantity stored in advance are stored. And the degree of similarity between the input state quantity and the pattern generated along the traveling direction of the material to be rolled, and the relationship between the certainty factor and the speed pattern for each of the plurality of stored patterns. And a membership function that defines the relationship between the certainty factor and the rolling down command pattern is stored in advance, and the obtained similarity is input, and the similarity is used as the certainty factor to determine the speed pattern and the speed from the stored membership function. This can be achieved by inferring the reduction command pattern and determining the speed of the electric motor of each rolling mill and the reduction amount of the reduction device based on the inferred speed pattern and the reduction command pattern.

Further, the above object is a heating furnace for heating a steel sheet conveyed by a conveying device by a plurality of burners, and detects from a plurality of detectors that detect a conveying speed at which the steel sheet is conveyed and a temperature in the furnace as a state quantity. A control unit for controlling fuel supplied to the burner based on the detected signal, an input unit for inputting a state quantity detected by the plurality of detectors, and an inside of the furnace that can be detected by the plurality of detectors. A plurality of patterns generated by combining at least one state quantity of the temperature or the conveying speed of the steel sheet along the traveling direction of the steel sheet is stored in advance, and the state quantity input from the input unit and the previously stored pattern are stored. A pattern recognition unit comprising: an output unit that obtains a similarity of each of the plurality of stored patterns and outputs a similarity to the plurality of patterns; For each of the plurality of patterns stored in the recognition means, a membership function defining the relationship between the degree of certainty and the combustion pattern of the burner is stored, and the similarity output from the pattern recognition means is input, and the similarity is output. And a command generation mechanism for inferring the combustion pattern of the burner from the stored membership function as a degree of certainty and determining fuel to be supplied to each burner based on the determined combustion pattern. can do.

[Action]

By combining a plurality of state quantities obtained from a plurality of control objects into a pattern, the entirety of the plurality of control objects is captured, the similarity of characteristic patterns included in the pattern is obtained, and the similarity of each characteristic pattern is determined. Using the similarity as the certainty factor, the operation amount of the control target is determined from a relationship in which the certainty factor and the operation amount for the control target are defined in advance. by this,
The amount of operation for optimizing the entire system can be determined, and the know-how of a skilled operator can be used as it is as knowledge, so that optimal control can be performed.

〔Example〕

FIG. 1 shows an embodiment of a control system according to the present invention.

The control target 1 includes a plurality of actuators, and the actuators are controlled by an actuator control device G6 that generates a control execution command for each actuator. The operation of the control target 1 and the plurality of actuators is performed by a detecting device G14 including various detectors disposed respectively.
Is detected by The plurality of detector output signals of the detecting device G14 are input to the optimality determining device G13. Optimal judgment device G
In 13, an input unit for recognizing the input detector output signals as a pattern, a storage function for storing the input pattern, and comparing the input pattern with a plurality of patterns previously stored are compared. It is composed of a processing unit that outputs a similarity to the previously stored pattern, and a command generation mechanism 12 that determines the operation amount of each actuator from the similarity and generates a command signal to the actuator control device G6. . In storing a pattern in the pattern storage function of the optimality determining device G13, an operator sets in advance an optimal output with respect to an input, activates the optimality determining mechanism, and outputs the best operation amount. And a display device G16 for changing the pattern storage content in advance when the control target 1 is changed or the actuator is changed.

On the other hand, the conventional system does not have the optimality judging device G13, and generally, the detecting device G14 is used for the actuator controller G
6 and the system was optimized and controlled for each actuator. Therefore, the optimality of the entire system has not been achieved, but by adding the optimality determination device G13 for determining the optimality of the entire system as in the present invention, the optimality of the entire system can be determined.
Even if one of the actuators fails, by this determination device, the action of the failed actuator can be covered by another actuator, or a change of the actuator,
This has the effect that it can flexibly respond to changes in the control target.

Next, a detailed description will be given using an example in which the control system of the present invention is applied to rolling mill control.

Hereinafter, an embodiment in which the present invention is applied to a rolling mill control system will be described with reference to FIG.

The rolling mill 1 to be controlled has a pair of work rolls 2 facing each other.
The rolled material 3 sandwiched between the work rolls 2 is so-called crushed by the rolling force acting between the work rolls 2 and the tension applied to the rolled material 3, and the rolled material is thinned by the pulling force to obtain a desired plate thickness. An intermediate roll 4 is interposed between the work rolls 2 and a back-up roll 5 is interposed between the intermediate rolls 4. A rolling force is applied to the back-up roll 5 by a rolling control function 6 using a force such as an oil pressure, and the rolling force is transmitted to the intermediate roll 4 via a contact surface between the back-up roll 5 and the intermediate roll 4. Then, the rolling force transmitted to the intermediate roll 4 passes through the contact surface between the intermediate roll 4 and the work roll 2 and between the work roll 2 and the rolled material 3, and
The rolled material 3 is plastically deformed by the rolling force to have a desired thickness.

By the way, the roll width of each of the work roll 2, the intermediate roll 4, and the backup roll 5 is wider than the plate width of the rolled material 3 and the roll is deformed because the rolling force is applied. For example, a portion of the work roll 2 outside the width of the rolled material is bent by the rolling force.

As a result, the end of the rolled material 3 is crushed to have a convex cross-sectional shape. In order to prevent this, a work roll bending force _FW is applied by a work roll bender 7 to the axis of the work roll 2 in a direction in which the interval is widened to prevent the end of the rolled material 3 from being crushed. Similarly, the intermediate roll bending is performed by the intermediate roll bender 8 on the shaft of the intermediate roll 4.
Add F _I.

Further, the intermediate roll shift 9 moves the intermediate roll 4 in the sheet width direction. By this movement, the force applied to the work roll 2, the intermediate roll 4, the backup roll 5, and the rolled material 3 is made asymmetric, thereby controlling the shape of the thickness of the rolled material.

On the other hand, the energy applied to the rolling mill to perform the rolling is not only spent for the plastic deformation of the rolled material 3 but also as sound, vibration and heat. The energy converted into heat is the rolled material 3
And the temperature of the work roll 2 is increased. Due to this temperature rise, the work roll expands and the roll diameter changes, but the roll diameter generally deforms unevenly. In order to uniformly control the roll diameter, a plurality of nozzles (not shown) arranged in the width direction of the plate and a coolant control mechanism 10 for applying a cooling liquid to the work roll 2 through the nozzles are provided. .

The shaft of the work roll 2 is connected to a speed control mechanism 11 including an electric motor or the like for moving the rolled material 3.

The control method for the rolling mill is the rolling control mechanism 6, work roll bender 7, intermediate roll bender 8, intermediate roll shift 9,
A command generation mechanism 12 for generating an operation command for an actuator such as a coolant control mechanism 10, a speed control mechanism 11, etc., and for the command generation mechanism 12, any one of a plurality of patterns in which the shape of the rolled material 3 is stored in advance. A pattern recognition mechanism 13 for determining whether the pattern belongs to a pattern, and outputting a certainty factor of the pattern; a shape detection mechanism 14 for detecting and outputting the thickness of the rolled material 3 for the pattern recognition mechanism 13; 14 and a storage mechanism for storing the output of the command generation mechanism 12
15 and a pattern recognition mechanism using the information of the storage mechanism 15
It comprises a learning mechanism 16 for changing 13 parameters by learning.

FIG. 3 shows a detailed view of the pattern recognition mechanism 13. Outputs of the shape detection mechanism 14 and the storage mechanism 15 are input to input cells 17, 18 of the pattern recognition mechanism 13, where the input signal is converted into a function value and output to the intermediate layer 19, The output of the input cell input to the layer 19 is the intermediate layer 19
Are input to cells 20 and 21 of FIG. Cell 20 at output of input cell 17
Input signal is input to the W ¹ ₁₁ times by the adder 24 with a weighting function 23, the output of the input cell 18 through a weighting function 26, is input to the adder 24, the adder 24 the weighting function Two
The outputs of 3 and 26 are added, input to the function unit 25, perform a linear or non-linear function operation in the function unit 25, and output to the intermediate layer 27 in the next stage. The cell 20 includes the weight functions 23 and 26, the adder 24, and the function unit 25.

Similarly, the cell 21 the output of the input cells 17 and 18 are input, the output of the input layer 17 is W ¹ ₂₁ doubling in weight function 28 adders
89, output to the next intermediate layer 27 via the function unit 90.

The intermediate layer 27 has the same structure as the intermediate layer 19, and the input cell
The output of the intermediate layer 19 is used in place of the outputs of 17,18.

Here, when the weights of the weight functions 23, 26, and 28 are represented by W ^k _ij ,
W ^k _ij indicates the weight applied to the i-th output of the (k-1) -th intermediate layer (when k = 1, the input cell) in the i-th cell of the k-th intermediate layer.

As described above, the signal input to the pattern recognition mechanism 13 passes through the input cells 17, 18 and the intermediate layers 19, 27, 29 of a plurality of stages, and is output in the form of removing the weight function and the adder from the cells of the intermediate layer. Output via layer 30. The input layer 31 is an input cell
Represents the sum of 17,18.

The feature of this pattern recognition mechanism 13 is that it requires only a simple product-sum operation, there is no repeated operation such as feedback, and each product-sum term in the intermediate layer is realized by hardware.
Since processing can be performed in parallel, high-speed operation is possible.

After the output layer 30 of the pattern recognition mechanism, a command value for each actuator can be stored in advance according to each output pattern, and a command value closest to the output pattern can be directly instructed to the actuator. In this method, the response is good, but the accuracy of the control is slightly lower than in the method described later.

Next, the processing result of the pattern recognition mechanism 13 is applied to the rolling mill 1 through the processing mechanism shown in FIG. That is, the output of the pattern recognition mechanism 13 is input to the operation amount determination means 32 provided in the command generation mechanism 12. The operation amount determining means 32 selects a processing mechanism most effective for processing an input signal from among a plurality of processing mechanisms prepared therein, executes the processing, and outputs the operation amount. Using the result of the manipulated variable determining means 32, the command value calculating means 33 generates specific command values of each actuator, for example, a rolling-down command to the rolling-down control mechanism 6, an intermediate roll bender instruction to the intermediate roll bender 8, and the like. The command generation mechanism 12 includes a pattern recognition mechanism.
Directly input the output of the shape detection mechanism 14 without going through 13,
It is also possible to perform processing by an optimal processing mechanism among the plurality of processing mechanisms prepared inside. However, in this case, when various inference mechanisms are used, it is necessary to enhance the knowledge base in order to sufficiently reflect the operation method of a specialized operator.

FIG. 5 shows the configuration of the manipulated variable determining means 32. The operation amount determination means 32 receives signals from the shape detection mechanism 14 and the pattern recognition mechanism 13 and activates the control mechanism 141. The control mechanism 141 uses the knowledge base 36 to determine the inference to be activated according to the type of problem. That is, the control mechanism
141 activates the production inference mechanism 142 when it is necessary to determine the cause by syllogism, and activates the fuzzy inference mechanism 143 when there is an ambiguous factor. Activates the frame inference mechanism 144, activates the semantic network inference mechanism 145 for problems where the causal relationship and the configuration of devices, etc. are network-like, and operates the diagnosis target in chronological order. For such a problem, the script inference mechanism 146 is activated. Further, the control mechanism 141 is an empirical problem that cannot be solved by the various inference mechanisms. The control mechanism 141 activates the optimization operation mechanism 111 for obtaining an optimum solution at high speed, can store the pattern in a pattern, extract features, and answer Feature extraction and answering mechanism 110 for solving problems that require a computer (consisting of Rumelhart-type Eurocomputer)
Start The processing result of the operation amount determination means 32 is
It is output to the command value calculation means 33 via 1.

FIG. 6 shows the structure of the knowledge base 36 which is the knowledge necessary for inference. The knowledge base is knowledge input from the outside based on the experience of a control expert, etc.The knowledge 106 is a production rule 147 for executing inference based on syllogism, and knowledge for performing inference based on ambiguous information. Fuzzy rule 148, frame of knowledge 149 that can be described in a certain framework such as the configuration of parts to be diagnosed, semantic network 150 that collects and organizes the relation between parts and common sense relations in the form of a network, , A script 151 for organizing and storing tasks in order to advance the tasks, and other knowledge 152 that cannot be described by the above-mentioned knowledge 147 to 151
And stored.

FIG. 7 is an explanatory diagram of the operation of the operation amount determining means 32. The processing of the control mechanism 141 is the pattern recognition mechanism 13, the shape detection mechanism 1
4, a processing step 200 for organizing the information from the storage mechanism 15 and converting it into data usable for the following processing, the above-mentioned step 2
Until the data prepared in 00 is exhausted, remove it in step 20
2, a repetition process step 201, an inference mechanism to be started from the information collected in the step 201 and a determination step 202 for determining a process, and various inference mechanisms 142 to 14.
6, feature extraction response mechanism 110, optimization operation mechanism 111, and P
It is composed of a general control mechanism 203 for executing algorithms of classical control such as ID control and modern control such as multivariable control, and an end processing step 204 for executing resetting of flags and the like necessary to end the above steps. Is done.

Here, the role of each processing mechanism will be described. The production inference mechanism 142 is suitable for control in which an expert of an operator assembles a logical establishment relationship using fragmentary production rules. The fuzzy inference mechanism 143 determines the amount of operation by quantifying the operator's ambiguous knowledge that can not be quantified, such as moving the actuator a little if the state of interest of the control target changes, so that the computer can process it. Suitable to do.

The frame inference mechanism 144 uses the knowledge of a frame that describes the relationship between the control devices, etc., and when the state of the control target of interest changes back to the original state, the operation is performed based on the relationship between those devices. This is suitable for determining the processing amount to perform for each related device.

Since the semantic net inference mechanism 145 organizes the frames, which are the fragmentary knowledge, and organizes the network, it is possible to obtain the influence of the operation result of the specific actuator, and to determine the compensation system. Suitable for assembling.

Since the script inference mechanism 146 infers based on procedural knowledge when a specific state occurs, it is suitable for sequence-type control in which a failure or the like must be dealt with by a predetermined procedure.

Also, the feature extraction answering mechanism 110 is configured to input the input patterns of the pattern recognition mechanism 13, the shape detection mechanism 14, and the storage mechanism 15 and the inference mechanisms 142 to 146 when the input patterns are input.
If the relationship between the outputs that are output is learned in advance, unlike inference mechanisms 142 to 146 that perform inference to determine the output,
The same effect can be output at high speed. The optimization operation mechanism 111 requires that the controlled object 1 usually has a strong nonlinearity, and if the operating point changes for any reason, the operation must be reset. In this case, the steepest slope method, dynamic programming, It is calculated by an algorithm such as linear programming, a hill-climbing method, a conjugate gradient method, or a Hopfield type neurocomputer, and performs an optimal response even for a nonlinear control target.

FIG. 8 is a diagram for explaining the operation of the production mechanism 142. The production inference mechanism 142 activated by the control mechanism 141 takes out the input processing 34 stored in the memory at the time of activation from the control mechanism 141 and the information stored in the input processing 34 one by one, and if the pattern information is stored in the memory, When there is no such information, an end determination mechanism 35 for ending the processing of the production inference mechanism 142 is executed. Using the type of pattern extracted by the end determination mechanism 35 and its certainty factor, the knowledge base 36
Then, in step 37, the type of the input pattern is compared with the premise of the rule. Using the comparison result, if step 38 matches, the next processing 39 is performed.
If they do not match, step 37 is executed. Step 39 replaces the input when matched with the conclusion of the rule.
The handling of the certainty at this time is based on the mini-Machus's theory, and is replaced by the minimum value or the maximum value before replacement. Step 40 executes step 41 when the conclusion of the replaced rule is an operation command, and executes step 37 for further inference when the conclusion does not match.

When the conclusion part is an operation command, the process 41 outputs the conclusion part and the certainty factor obtained in the processing steps to the command value calculating means 33.

FIG. 9 shows the command calculation means 33. Command value calculation means 33
Determines whether or not the command as the inference result obtained by the manipulated variable determining means 32 and the memory 42 for storing the certainty thereof have been processed, and if the command has been processed, the command value calculating means Step 43 for terminating 33, if not processed, actuators 7, 8, 9, 10, 1 such as reduction control mechanism 6
Each command is taken out and the center of gravity of the operation amount is calculated based on the degree and certainty of the actuator operation obtained by various inferences.
Processing 44 for collecting the center of gravity of the operation amount of the same actuator, obtaining a new center of gravity, and setting it as a command of the corresponding actuator 44
Consists of

Various inferences can be made by obtaining such a command value calculation means 33.
142 to 146, the feature extraction circuit mechanism 110, the optimization operation mechanism 111, and the general control mechanism 203 have a feature that commands to the actuators individually obtained can be handled in a unified manner.

FIG. 10 shows the configuration of the input switching device 125 necessary for the learning. The input switching device 125 uses a switch mechanism 156 controlled by a learning mechanism, and
One of the output of the learning mechanism 16 and the output of the learning mechanism 16 is output to the input layer 31. The state of the switch mechanism 156 in FIG. 10 indicates a state in which learning is performed.

FIG. 11 shows the configuration of the learning mechanism 16. The learning mechanism 16 includes an input pattern generating mechanism 45, an output pattern generating mechanism 47, an output matching mechanism 46, and a learning control mechanism 48. The output matching mechanism 46 outputs a command from the output layer 30 to a command generation mechanism.
Output o ₁ of the distributor 139 for outputting the to abutting mechanism 46 and 12, o _i, o _n the output o _T1, o _Ti output pattern generating mechanism 47,
The adder 161, 162, 163 a difference between the o _Tn, the deviation e _1, e _j, calculated as e _n, and outputs to the learning control mechanism 48. Distributor 139
The output o _1, o _j, o _n of generated by the output of the input pattern generating mechanism 45 is input to the input layer 31 of the pattern recognition mechanism 13 ((Rumelhart type two euro-computer). At this time, the input pattern The generating mechanism 45 and the output pattern generating mechanism 47 are controlled by the learning control mechanism 48.

FIG. 12 shows the relationship between the weight function W _ij 23 and the learning control mechanism 48 in the learning process. Upon receiving the deviation e _k which is the output of the adder 161, the learning control mechanism 48
The value of the weight function W _ij 23 of the cell 20 constituting the said deviation is changed to the decreasing direction.

FIG. 13 shows a processing outline 170 of the learning control mechanism 48.
When the learning mechanism 16 is activated, the processing 170 of the learning control mechanism 48
Is started. The processing 170 activates the input pattern generating mechanism 45 and the output pattern generating mechanism 47, and inputs a teacher signal, a preprocessing 171 for generating a desired output, and the deviation e _k
Or the following steps 173, 174 and 175 are repeated until the sum of the squared deviations is within the allowable range, a step 174 for sequentially extracting intermediate layers of interest from the intermediate layer close to the output layer 30 toward the input layer 31, Steps 174 for sequentially extracting a cell of interest in the intermediate layer, steps 175 for changing the weight function W _ij 23 of the cells extracted in the direction in which the deviation e _k becomes smaller, and steps 176 for ending the learning process. Be composed.

By providing such a learning mechanism, a new phenomenon that has not been taken into consideration occurs, and if a countermeasure for it is determined, the knowledge can be reflected.

FIG. 14 shows the configuration of the storage mechanism 15 of FIG. The storage mechanism 15 has a command generation mechanism 12, a memory element 49 to which the outputs of the shape detection mechanism 14 are input, a memory element 50 in which the contents of the memory element 49 are transferred after a certain period of time, and data is sequentially transferred to the memory element. Memory element 51 reached after a specific time has elapsed
The contents of each of the memory elements 49, 50, and 51 are applied to the pattern recognition mechanism 13 and the learning mechanism 16 via an arithmetic mechanism 510 for performing differentiation, integration, and the like of the pattern.

This storage mechanism 15 enables operations such as differentiation and integration, which can take into account temporal changes in the shape detection mechanism 14 and the command generation mechanism 12, for example.

FIG. 15 shows a mechanism for recognizing a pattern using an input in the vicinity of the nozzle, since the effect of the nozzle in the coolant control affects only a certain length from the position of the nozzle. The output of the shape detection mechanism 14 is input to the memory 52 of the pattern recognition mechanism 13, the signal input to the memory 52 is input to the memory element 54 via the gate circuit 53, and the signal input to the memory element 54 is When the gate circuits 53 and 56 are turned off, the gate circuit 55 is turned on, and in synchronization with the clock, the information of the memory element 54 is transferred to the memory 57 and for a certain period of time. Over time, the signal on memory element 54 reaches memory element 58, the signal on memory element 57 reaches memory element 54, and in the next clock, memory elements 54, 57, 58
Signal goes round, gates 53 and 56 are turned on, and gate 55
Is turned off, the contents of the memory element 54 are stored in the memory element 59, and the information of the memory elements 54, 57, 59 is input to the input layer 31.

By providing such a memory 52, there is an effect that the number of cells in the input layer 31, the intermediate layers 19, 27, 29, and the output layer 30 of the pattern recognition mechanism 13 can be significantly reduced.

FIG. 16 shows an example in which the controlled object simulator 60 is used for the input pattern generation mechanism 45 and the output pattern generation mechanism 47 of the learning mechanism 16.

The shape pattern generated by the operation or data of the operator in the output pattern generating mechanism 47 has the same function as the command generating mechanism 12 in FIG. 2, and is input to the command generating mechanism 12 provided separately in the learning mechanism. The command generation mechanism 12 generates commands for various actuators according to the pattern, and the commands are input to a control target simulator 60 provided in the input pattern generation mechanism 45, and the various actuators 6, 7, 8 , 9, 10, 11 and the operation of the rolling mill 1 are simulated, and when the response is poor, the command generation mechanism 12 and the parameter adjustment mechanism 61 for changing the parameters of the control target simulator 60 are used. Is adjusted so as to have a desired shape, and is used as an input to the pattern recognition mechanism 13.

The operation of the control method having the configuration described above will be described below using a specific example.

Initially, the initial value of the weight function W _ij 28 of the intermediate layer 19, 27, 29 of the neuro-computer constituting the pattern recognition mechanism 13 is initially a random number or an appropriate value, for example, a value that can be taken by a weight function (0 to 1.0). Then set it to half (0.5). At this time, for example, even if the concave rolling mill shape pattern generated by the input pattern generating mechanism 45 of FIG. 17 is input, the output of the output line 70 that is concave in the output of the output layer 30 does not become 1, In addition, the probability that the output line 71 of the output layer 30 is a convex output is not zero.

Therefore, the output line 72 of the output pattern generating mechanism 47 of the learning mechanism 16 corresponding to the output line 70 of the output layer 30 outputs 1 and the output of the output line 73 of the output pattern generating mechanism 47 corresponding to the output line 71 outputs zero. . Upon receiving these outputs, the output matching mechanism 46 receives the ideal output (output pattern generating mechanism 47) and the deviation of the output of the pattern recognition mechanism 13, and the learning control mechanism 48 receives the weight function _Wij of the pattern recognition mechanism 13. Is changed in the direction in which the deviation decreases in proportion to the magnitude of the deviation. A steepest gradient method is a typical example of this algorithm.

Accordance connexion to the processing of FIG. 13, to change the weight of the sequential weighting function, the square sum of e _k of Figure 12 falls within the allowable range, the operation of the learning device 16 is completed.

When the same waveform as the output pattern of the input pattern generation mechanism 45 in FIG. 17 is input from the shape detection mechanism 14 in FIG. 2 after the learning is completed, the pattern recognition mechanism 13 outputs 1 from the output line 70 of the output layer 30. Then, zero is output from the output line 71 of the output layer 30.

Next, when the waveform shown in FIG. 18 called a convex type is input and learning is not completed, the output of the output line 71 representing the convex type of the pattern recognition mechanism 13 is 1, and Does not become a pattern in which the output line 70 becomes zero. Therefore, as described above, a typical convex pattern is used as an input signal,
The output of the output pattern generating mechanism 47 is set so that the values corresponding to the outputs of the output lines 71 and 70 are 1 and 0, respectively. Learning mechanism 16
Changes the weight function W _ij , and when learning is completed,
When the convex waveform of FIG. 18 is input to the pattern recognition mechanism 13, the output line 71 of the output layer 30 of FIG. 17 becomes 1 and the output line 70 becomes zero.

As a result, the waveform of FIG.
The output is output from the output layer 30 as a certainty factor of 40% by the output line 71 indicating the convex waveform input in advance as described above. At the same time, the output line 70 indicating a concave waveform
Is output as 50% confidence.

FIG. 20 shows a rolled material shape in consideration of a temporal change of the rolled material. State immediately below the rolling mill work roll 2 is t _0, the value at that time is x _0. Assuming that the sampling period of the calculator is T ₀ , the height of the sheet thickness at time t ₁ before T ₀ seconds is x ₁ , T ₀ x ₂
The height of the plate thickness at the time t ₂ before the _second is x ₂ ,.

That is, at time t _2, the height x ₂ are input to the storage device 15,
It is stored in the memory element 49 of FIG. When the height x _{1 of the} next sampling time t ₁ is input to the storage mechanism 15,
At that time, the data x ₂ of the memory element 49 is the memory element
Together are transferred to 50, the contents of the memory element 49 is rewritten to x _1.

On the other hand, the operation mechanism 510 performs various operations using the contents of the memory elements 49 and 50. For example, when a differential value is required, (x ₂ −x ₁ ) / T ₀ , and when an integrator is required, (x ₁ + x ₂ )
What is necessary is just to execute the calculation which becomes × T ₀ . That is, since the differentiator can determine the change speed of the shape, the pattern recognition mechanism 13 can improve the response to the change.

On the other hand, the integrator can exhibit features such as a function of removing noise and the like.

The pattern recognition mechanism 13 can have functions such as a differentiator, an integrator, and a proportional element not including a temporal element.

Furthermore, the data stored in the storage mechanism 15 is also required,
It can be used for the input pattern generation mechanism 45 used for learning.

By the way, as shown in FIG. 29, the plate thickness of the rolled material in the roll axis direction of the rolling mill at the time point t ₀ is x ⁰ ₀ , x ⁰ ₁ ,... X ⁰ _n-1 , x ⁰
and _n, the at the same position T ₀ (sampling period) before the plate thickness of the ^{_{state, x 1 0, x 1 1}} , ... When ^{_{x 1 n-1, x 1}} n, in some point t _0, the Husband x ⁰ _n , x ⁰ is added to memory element 54,57,58 in Fig.14
_n-1, ... x ^{₀ 0} is stored. The memory element 59, since Gyotsu the same operation as the memory mechanism 15 described above, T ₀ x ¹ ₀ point is before the data at the time ^{_{t 1, x 1 1, ...}} x 1 n memory elements 59 other Is stored in

FIG. 22 shows an example of the production rule or the fuzzy rule. (Corresponding to the production rule 147 and the fuzzy rule 148 in FIG. 6).

When the pattern recognition mechanism 13 obtains an output with a confidence of 50% for the concave type, it is checked against the premise of the production rule and matches the concave rule 80. As a result, a rule 81 that weakens the vendor (to a small extent) is obtained. On the other hand, with a convex certainty factor of 40%, it matches with the premise 82, and as a result, the vendor is strengthened (large degree).

As a result, as shown in FIG. 23, the command generation mechanism 12 is represented by the area of the shaded area of B because the operation amount of the vendor is 50% convex certainty as a result of the comparison with the rule. On the other hand, since the certainty factor of the concave type is 40% and the certainty factor S is 40%, the area of the hatched portion of S in FIG. 22 is the area. Next, in the command generating mechanism 12, 65%, which is the value of the center of gravity C obtained by combining the centers of gravity A and B in the hatched portion, becomes the operation amount of the vendor.

Next, as shown in FIG. 15, when the influence of the actuator is local unlike the vendor or the shifter as in the coolant control, the waveform of FIG. 24 (a) is stored in the memory elements 54, 57, 58.
To memorize. A part of the waveform stored in the memory element (24th
The part (a) in the figure (a) is processed by the pattern recognition mechanism 13 and the command generation mechanism 12, and by controlling one nozzle A of the coolant control mechanism 10, the amount of the coolant is controlled, and the roll is flattened. You do it.

Now, both sides of x ⁰ _{n-1 in} FIG. 21 corresponding to nozzle A
x ⁰ _n, when compared with the value of x ⁰ _n-2, the larger the x ⁰ _n-1, the conclusion that the central portion size 85 is obtained from Figure 15. On the other hand, x ⁰
If _n-1, x as a relation of _{^{x 1 n-1 0 n-}} 1, x 1 n-1 positive is, x ⁰
_{Since n-1} tends to increase, the derivative becomes positive, and
And turns on the coolant as a result. The degree is large (B). As a result, x ⁰ _n-1 and x ⁰ _n-1 hardly change.

When the control of the nozzle A is completed, the memory elements 54, 57,
The contents of 58 and 59 are shifted one by one. As a result, as for the waveform input to the pattern recognition mechanism 13, the area indicated by b in FIG. 24 (a) is input, and the pattern recognition mechanism 13 and the command generation mechanism 12 control one nozzle of the coolant control mechanism 10. B is controlled.

When such processing is performed, the waveform starts from pattern A in FIG. 24 (a), reaches pattern B, and further shifts the memory contents, whereby the waveform in FIG. 24 (a) is reproduced in memory 52. The content of the memory element 54 shown in FIG. 15 is stored in the memory 59 after a predetermined time has elapsed since the last time the pattern shown in FIG.
Then, the waveform of the shape detection mechanism 14 is stored in the memory element 54.

Further, when the arithmetic mechanism 510 shown in FIG. 14 is provided between the memory 52 and the input layer 31, it is obvious from FIG.

Next, a method of learning a pattern serving as a reference for the pattern recognition mechanism 13 will be described.

The waveforms 62 and 63 in FIG. 18 are generated by the input pattern generation mechanism 45 in FIG. This pattern
The data is written in the memory of the input pattern generation mechanism 45 or the pattern stored in the storage mechanism 15 of FIG. 2 is used. The signal input to the input layer passes through the intermediate layers 19,.
Appears as output from. At this time, the weight function W ^k _{ij of the hidden} layer
Is an initial value. From the output pattern generation mechanism 47, a pattern desired to be output from the pattern recognition mechanism 13 corresponding to the output of the input pattern generation mechanism 45 (for example, the input pattern generation mechanism 45 is a standard pattern, When one output terminal of the output layer 30 is assigned to the standard pattern, a pattern in which the assigned output terminal becomes 1 and the other terminals become zero is input to the output matching mechanism 46. When the learning is not completed, the output pattern of the output layer 30 and the waveform of the output pattern generation mechanism 47 are different. As a result, the output of the output matching mechanism 46 outputs an output according to the degree of the pattern difference. By calculating the mean square of this value and the deviation, the power spectrum of the deviation and the like can be obtained. According to the deviation, the weight function W ^k _ij is changed from the intermediate layer 27 near the output layer to the intermediate layer 19 near the input layer 31 in order. Various methods can be used to change the function W ^k _ij . For example, the steepest slope method is used in an optimization problem of minimizing the deviation.
As a specific method, the weight function W ^k _ij of interest is slightly changed in the upward direction, and as a result, the weight function W ^k _ij is moved in the decreasing direction while looking at the direction in which the deviation value changes,
The movement amount is large when the change in the deviation value is small, and is small when the change in the deviation value is large. Or
When the variation of the weight function W ^k _ij of the intermediate layer 19 closest to the input layer 31 is completed, the deviation value of the output matching mechanism 46 is checked again, and learning is terminated when the value falls within the allowable error range. .

This control is performed by the learning control mechanism 48. Grace, a pattern recognition mechanism that uses this learned result for pattern discrimination
13 does not clarify the operation of why the pattern can be identified and why the learning works, but the number of weight functions is larger than the number of inputs and outputs, and there is a degree of freedom in the value. It is said that a good recognition result can be obtained even if the value is slightly out of order or a large number of patterns are stored.

On the other hand, it is very difficult to determine what pattern should be used for the input pattern generation mechanism 45 and the output pattern generation mechanism 47. Fortunately, the theory of system identification has been established in the field of control theory as a method for accurately deriving a model when the operation of the control target 1 is operated near a certain operating point. However, in the entire operation region, it is difficult to model an object having strong nonlinearity.

Therefore, a model is created in a specific operation area, control is performed, and the relationship between the input and response of the control system that works well in that state is determined by simulation, and this is used as learning data.
In this procedure, the operating point is sequentially moved in the entire operation area of the control system, and the optimal modeling and control command at each time are obtained and learned. That is, the parameters of the controlled object simulator 60 in FIG. 16 are adjusted, and the controlled object simulator 60 is operated accurately at a specific operating point. Then, the input pattern generation mechanism 45, so that the control target generates a typical pattern
The parameter adjustment mechanism 61, the control target simulator 60, and the command generation mechanism 12 are operated, and the outputs of the input pattern generation mechanism 45 and the control target simulator 60 are used as the output pattern and input pattern of the learning mechanism 16, respectively.

In the control method having such a configuration, the target waveform is abstracted by the pattern recognition mechanism, and the control including the ambiguity can be performed by the control mechanism.

Next, FIG. 25 shows an embodiment in which the present invention is applied to a hot rolling mill. Hot rolling system with rolling stands 500,50
1,502,503,504, electric or hydraulic pressure reduction device 505,506,507,
508,509, electric motor 510,511,512,513,514, looper 515,516,
517,518,519, tension detectors 520,521,522,523,524, rolled material 525, input / output interface 526, pattern recognition mechanism
13, comprising a control mechanism 528. Rolling stands 502,503
25 is expanded as shown in the lower part of FIG. 25. For example, the rolling stand 502 includes work rolls 529 and 530, intermediate rolls 531 and 532, and backup rolls 533 and 534. The force applied by the hydraulic pressure reduction device 507 includes backup rolls 533,534, intermediate rolls 531,532, and work rolls.
It is added to the material to be rolled 525 through 529, 530, and the thickness thereof is controlled. Also, it comes into contact with the material to be rolled through a roll,
By controlling the rotation angle of the attached electric motor 53, the looper 517 controls the tension applied to the material to be rolled 525.

Further, a tension meter 522 contacts the rolled material 525, and measures the tension of the rolled material 525. The input / output interface has the tension of the output of the tension meter, the output load of the load meter attached to the hydraulic pressure reduction device, the speed of the motors 510 to 514 connected to the intermediate roll and driving the roll, the position of the looper, and the current of each motor. A value or the like is input to the input / output interface. The state of the control target input to the input / output interface 526 is input to the pattern recognition mechanism 13 composed of a neural network. A pattern (distribution) of various signals input to the pattern control mechanism 13, for example, a tension distribution composed of a plurality of tensiometer signals is shunted to any of the pre-stored patterns, and how many its components are included. Is extracted as a similarity and output to the control mechanism 528. The control mechanism receives the similarity as a certainty factor, performs fuzzy arithmetic, determines a command for each actuator, and sends the command to the input / output interface.

The above operation will be described with reference to FIG. For example, the tension patterns of the output patterns of the plurality of tension detectors 520 to 524 are input to the neural network 527 as the pattern recognition mechanism 13 via the input / output interface 526.
At the same time, a load pattern, a speed pattern created by the speeds of the motors 510 to 514, and a rolling stand (not shown)
A temperature pattern obtained from a thermometer that measures the temperature of the material to be rolled, which is installed anywhere between 500 and 504, a rotation angle pattern obtained from the rotation angle of the looper, a current pattern of the electric motor, and the like are input.

The neural network 527 outputs, as a similarity, a ratio of a typical pattern component learned in advance from each pattern to the control mechanism 528 including an inference mechanism.

Next, the operation of the control mechanism 528 will be described with reference to FIG. For example, as an operator's know-how, as shown in the knowledge base, "If the tension pattern is falling to the right, the load pattern is concave, the speed pattern is convex, and if the temperature pattern sharply falls to the right, the speed pattern is Flat,
The rolling command pattern is set to lower right. In some cases, the neural network 527 is convinced that the similarity of the tension pattern to the right, the concave similarity of the load pattern, the convex similarity of the speed pattern, and the similarity of the temperature pattern to the rapid downward slope are each determined. Get as a degree.

The inference processing 536 performs fuzzy inference based on the knowledge base 537 and the plurality of degrees of certainty, and leads to a down-draft command having a downward-sloping pattern. In the inference processing, the condition part uses a method that determines the degree of conformity with the rule by mini-max operation and generates an actuator command at the center of gravity of the conclusion part. In this case, it is assumed that the command has concluded that the rolling-down command pattern falls to the right. In response to the conclusion, the command generation mechanism 538 generates a specific individual hydraulic pressure reduction device pressure reduction command from the downward-sloping pattern. Each command is input to the hydraulic pressure reduction device of the actuator.

FIG. 28 shows a configuration diagram when the configuration of FIGS. 25 to 27 of the present invention is applied to a heating furnace. Here, only differences from FIGS. 25 to 27 will be described. The billet 540 is sent to the heating furnace main body 542 by the feed pitch control device 541. In the heating furnace body, a fuel control device 544 for controlling the combustion of the burner 543,
Blower fan controller 545 for sending air for combustion, and temperature distribution detection mechanism for measuring the temperature distribution in the furnace
546. In the case of a heating furnace, temperature distribution (temperature pattern) and speed distribution (pattern) of the billet feeder
It is self-evident that the same control as in the hot rolling system can be performed after the above is input to the neural network.

Further, an application example of the present invention will be described by taking continuous casting as an example. FIG. 29 shows a continuous casting facility. The melted steel material 550 supplied from the converter becomes a steel material while being cooled by the rolls between the plurality of rolls. Although an electric motor is connected to one roll in FIG. 29, an electric motor 552 is actually connected to each roll 551. Although the surface temperature of the steel material 550 is not shown directly, the surface temperature distribution (pattern) of the steel material 550 is measured by a plurality of temperature detectors and input to the input / output interface. The electromagnetic stirrer 553 stirs by electromagnetic force in a state where the steel material is melted, and makes the material of the steel material constant. In this case, a temperature pattern and a speed pattern of the plurality of rolls 551 are detected and controlled. The pattern detection and control method is the same as in hot rolling.

The present invention can be applied to a process line for a blast furnace, a plating bath, or the like.

Although the embodiment has been described using a steel system as a specific example of the present invention, the control target 1, various actuators
6,7,8,9,10,11 need not be limited to the rolling mill system, and it is obvious that they can be applied to general controlled objects, actuators and controllers. For example, the present invention can be used for controlling a system such as a railway operation management system that recognizes a train schedule pattern and returns a delayed train to a normal schedule according to various schedule change rules. That is, the train operation is represented by a diagram, and the pattern recognition mechanism 13 extracts delayed features. Next, based on the feature amount, the inference mechanism creates a diagram using various rules such as, for example, passing trains at a station. In response to the result of the inference mechanism, the command calculation mechanism 33 generates an operation command for an individual train. Actuator. The train operates according to the command.

〔The invention's effect〕

According to the present invention, by combining a plurality of state quantities obtained from a plurality of control targets into a pattern, the entire control target is captured, and the similarity of a characteristic pattern included in the pattern is obtained. Using the similarity of each characteristic pattern as a certainty factor, the operation amount of the control target is determined from a function in which the certainty factor and the operation amount for the control target are defined in advance. As a result, the amount of operation for optimizing the entire system can be determined, and the know-how of a skilled operator can be directly used as knowledge, so that optimal control can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram of one embodiment of the present invention, FIG. 2 is an embodiment in which the present invention is applied to a rolling mill control system, FIG. 3 is a pattern recognition block diagram, FIG. 5 is a block diagram of the operation amount determining means, FIG. 6 is a knowledge base block diagram, FIG.
FIG. 8 is an explanatory diagram of the operation of the operation amount determining means, FIG. 8 is an explanatory diagram of the operation of the production mechanism, FIG. 9 is a configuration diagram of the command value calculation device, FIG. 10 is a configuration of the input switching device, and FIG. Structure of the mechanism, FIG. 12 is a diagram showing the relationship between the learning control mechanism and the weight function of the node, FIG. 13 is a basic processing diagram of the learning control mechanism, FIG. 14 is a configuration diagram of the storage configuration, and FIG. 15 is a pattern recognition mechanism. Fig. 16,
The figure shows the configuration when the learning mechanism is equipped with a simulator.
FIG. 18 is an explanatory diagram of the operation of the pattern recognition mechanism, FIG. 18 is an example of an input pattern, FIG.
Fig. 20, Fig. 21 is an explanatory diagram of a temporal change of a rolled material, Fig. 22 is a diagram showing an example of a production rule and a fuzzy rule, Fig. 23 is an explanatory diagram of a method of converting a similarity into an operation amount,
FIG. 24 is an explanatory diagram of the processing status of an input waveform, FIG. 25 is a configuration diagram of a hot rolling system, FIG. 26 is a specific example of a feature extracting unit of the hot rolling system, and FIG. 27 is an operation diagram of a control function FIG. 28 is a block diagram of the heating furnace, and FIG. 29 is a block diagram of the continuous casting. 1 ... controlled object, 13 ... pattern recognition mechanism, 12 ... command generation mechanism, 16 ... learning mechanism, 19, 27, 29 ... middle layer, 30 ...
... output layer, 31 ... input layer.

──────────────────────────────────────────────────の Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI Technical indication location C21D 9/00 101 G05B 13/02 Q 9154-4E B21B 37/00 116B F27B 9/40 9154-4E 116P G05B 13/02 9154-4E 137A 9154-4E BBK 9154-4EZ Z (72) Inventor Masaaki Nakajima 5-2-1 Omikacho, Hitachi City, Ibaraki Pref. Hitachi, Ltd. Omika Plant (56) References JP JP-A-61-180304 (JP, A) JP-A-2-1577981 (JP, A) JP-A-2-170906 (JP, A) JP-A-1-183703 (JP, A) JP-A-2-165201 (JP) JP-A-3-266007 (JP, A) JP-A-2-244203 (JP, A) JP-A-4-90001 (JP, A) JP-A-4-70906 (JP, A) 4-9002 (J P, A) JP-A-2-165388 (JP, A) Hitachi Review 71 [8] (1-8) Toshio Sakai and 5 others "Shape control by fuzzy theory" 803-808

Claims (3)

(57) [Claims] A speed of an electric motor driving a roll of a rolling mill, which is detected as a state quantity of an actuator from a detector provided in at least each rolling mill of a rolling system for rolling a material to be rolled by a plurality of rolling mills, and At least one state quantity among the load of the rolling-down device for controlling the thickness of the rolled material and the temperature and tension of the material to be rolled detected as the state quantity of the material to be rolled detected from the detector provided on the output side of each rolling mill. A control device that controls the amount of reduction of the reduction device and the speed of the electric motor that each of the plurality of rolling mills has based on the following. The detected speed of the electric motor, the load of the reduction device, the temperature of the material to be rolled, and the material to be rolled The input unit for inputting at least one state quantity among the tensions from a detector provided on each rolling mill or a detector provided on the output side of each rolling mill, and the state quantity input from the input unit A plurality of generated patterns are stored in advance, and the degree of similarity between a pattern generated by combining the input state quantities along the traveling direction of the material to be rolled and the previously stored pattern is stored in the plurality of stored patterns. Pattern recognizing means comprising: an output unit that obtains each of the patterns and outputs a similarity to each of the plurality of stored patterns; and a certainty factor and a speed for each of the plurality of patterns stored in the pattern recognizing means. A membership function that defines a relationship between the pattern and a certainty factor and a reduction command pattern is stored, and the similarity output by the pattern recognition means is input, and the stored member is used as the certainty factor for the similarity. A speed pattern and a rolling command pattern are inferred from the ship function, and the inferred speed pattern and the rolling command pattern are deduced. And a command generation mechanism for determining the speed of the electric motor of each of the rolling mills and the amount of reduction of the reduction device based on the following. 2. The speed of an electric motor driving a roll of a rolling mill, which is detected as a state quantity of an actuator from at least a detector provided in each rolling mill of a rolling system for rolling a material to be rolled by a plurality of rolling mills, and At least one state quantity among the load of the rolling-down device for controlling the thickness of the rolled material and the temperature and tension of the material to be rolled detected as the state quantity of the material to be rolled detected from the detector provided on the output side of each rolling mill. In a control method of controlling the amount of reduction of the reduction device and the speed of the electric motor that each of the plurality of rolling mills has based on, the speed of the electric motor, the load of the reduction device, the temperature of the material to be rolled, and the tension of the material to be rolled. At least one of the state quantities is input from a detector provided in each of the rolling mills or a detector provided on the output side of each rolling mill, and a plurality of parameters generated for the input state quantities stored in advance are stored. And the degree of similarity between the input state quantity and the pattern generated along the traveling direction of the material to be rolled, and the relationship between the certainty factor and the speed pattern for each of the plurality of stored patterns. And a membership function that defines the relationship between the certainty factor and the reduction command pattern is stored in advance, and the obtained similarity is input, and the similarity is used as the certainty factor to calculate the speed pattern and the speed from the stored membership function. A control method comprising: inferring a reduction command pattern; and determining a speed of a motor and a reduction amount of a reduction device of each rolling mill based on the inferred speed pattern and the reduction command pattern. 3. A heating furnace for heating a steel sheet conveyed by a conveying device by a plurality of burners, comprising a plurality of detectors for detecting a conveying speed at which the steel sheet is conveyed and a temperature in the furnace as a state quantity. A control device for controlling fuel supplied to the burner based on a detected detection signal, comprising: an input unit for inputting a state quantity detected by the plurality of detectors; and an inside of the furnace that can be detected by the plurality of detectors. A plurality of patterns generated by combining at least one state quantity of the temperature or the conveyance speed of the steel sheet along the traveling direction of the steel sheet are stored in advance, and the state quantity input from the input unit and the previously stored pattern are stored. A pattern recognizing unit comprising: an output unit that obtains a similarity between the plurality of stored patterns for each of the stored patterns; and an output unit that outputs a similarity to the plurality of patterns. For each of the plurality of patterns stored in the recognition means, a membership function defining the relationship between the degree of certainty and the combustion pattern of the burner is stored, and the similarity output from the pattern recognition means is input, and the similarity is output. A command generation mechanism for inferring a combustion pattern of the burner from the stored membership function as a degree of certainty and determining fuel to be supplied to each burner based on the determined combustion pattern. Characteristic control device.